Hi all
One should be aware of some stats about twinning (and in general):
1) If you have perfect hemihedral twinning and no correlated
rotational NCS then random R factors are around 40 instead of
Luzzati's 58.
2) If you have perfecthemihedral twinning and you are not modelling
twin then random R factor is around 50 (again instead of 58). It may
confuse things after molecular replacement.
3) In twinning case reflections are related with each other. refmac
groups them together so that Rfree is unbiased by twinning. However
after non-twin refinement if you start twin refinement then Rfree in
the first few cycles can be small. If you use consistenlty refmac
with twin refinement then Rfree should be fine
Intuitive reason for the first two phenomena is that if the crystal
is twinned then the distrbution of the data becomes narrower and
properties like Rfactor that are dependent on the spread of the
distribution become smaller. (There is a mathematical formulation
that gives the exact numbers also). Presence of pseudo translation
(broadens the distribution) and rotation (data points are not
independent) complicates things a bit further.
The reason for the third phenomenon should e clear.
Unfortunately statistics like Rfactor are not robust to violations of
such assumptions. You can get very interesting Rfactors under
different assumptions.
For me the best indicator is the electron density (do you see new
features and can you remove wrong bits) and chemical sensibility of
the electron density you see.
Garib
On 14 Aug 2008, at 07:46, Jan Abendroth wrote:
Hi all,
kind of a weird problem - the R-factors of a refinement using the
new twin refinement in refmac are low, almost suspiciously low:
A good 1.9AA data set, space group H3/R3, many statistics starting
with truncate's cumulative intensity distribution clearly suggest
twinning. The structure (SSGCID target) was solved by MR, very
rigid beta-helix (see 3bxy, very cool fold!), maps nice. I use
refmac 5.5.0046 for refinement, with and without the new TWIN flag.
Here my concerns:
Despite 0.3/0.7 twin domains as suggested by various programs and
refined by refmac, the difference of between the Rs of the twinned
refinement and the non-twinned refinement are constantly rather
little: 0.143/0.174 for the twinned case, 0.218/0.275 for the non-
twinned case. This seems rather low to me for a quite high twin
fraction. Plus the R-factors for the twinned case are really low
for a 1.9AA structure. As expected, the maps from the twin
treatment are a bit nicer that for the non-twinned treatment.
Any reasons for concerns or just a very rigid structure that
refines really well or refmac handling twinning really nicely?
Thanks for any input
Jan
--
Jan Abendroth
deCODE biostructures
Seattle / Bainbridge Island WA, USA
work: JAbendroth_at_decode.is
home: Jan.Abendroth_at_gmail.com
